Breaking AI: A Practical Guide to Testing LLM Applications and Agents

Breaking AI: A Practical Guide to Testing LLM Applications and Agents is a comprehensive, hands-on guide to understanding and securing the rapidly evolving world of artificial intelligence.

Written by Divine Chukwu, this book takes security professionals, penetration testers, developers, and AI enthusiasts beyond the chatbot interface to explore how modern AI systems are actually built, where they fail, and how to test them responsibly.

You'll learn how to assess prompt injection, system prompt exposure, guardrail weaknesses, RAG and retrieval attacks, vector database security, cross-tenant failures, AI memory risks, tool and function-calling abuse, excessive agency, autonomous agents, multi-agent systems, sandbox escapes, and supply-chain risks.

The book also provides a practical methodology for conducting professional AI security assessments—from scoping and threat modelling to testing, evidence collection, and writing security reports.

With a dedicated AI Test-Case Library, Engagement Checklist, Sample Reports, and Glossary, Breaking AI gives you a structured foundation for entering the growing field of AI security and AI red teaming.

AI security is not just about the model. It's about the prompts, data, tools, agents, integrations, and systems that trust its output. This book teaches you how to test the entire AI stack.

By Divine Chukwu | Devwalls Security — Offensive Security for the AI Stack